US-12620476-B2 - Cardiac image workflow interpretation time prediction

US12620476B2US 12620476 B2US12620476 B2US 12620476B2US-12620476-B2

Abstract

A method predicts an interpretation time for a medical image examination of a subject comprising one or more medical images. A plurality of data inputs is obtained, where the data inputs are associated with the medical image examination or the subject of the medical image examination, and the data points represent parameters affecting the interpretation time. The plurality of data inputs are input to a trained artificial intelligence algorithm, wherein the algorithm automatically provides a predicted interpretation time based on said plurality of data inputs. The predicted interpretation time is output to a clinical management system. A clinical management system incorporating the aforementioned method and a computer program product encoded with the aforementioned method are also provided.

Inventors

Ali Sadeghi
Lucas de Melo Oliveira
Deyu Sun
Hua Xie
Claudia ERRICO
Jochen Kruecker

Assignees

KONINKLIJKE PHILIPS N.V.

Dates

Publication Date: 20260505
Application Date: 20220118

Claims (17)

1 . A method implemented by a processor for optimizing prediction of interpretation times for a plurality of medical image examinations of subjects, each medical image examination comprising one or more medical images, the method comprising: training a regression-based, artificial intelligence algorithm by obtaining training data inputs and corresponding actual interpretation times for a plurality of examinations, training the artificial intelligence algorithm to map states defined by the training data inputs to corresponding predicted interpretation times using a reinforcement learning approach, and minimizing a reward function to optimize the artificial intelligence algorithm to improve the predicted interpretation times over time, wherein the reward function comprises an absolute value of a difference between the predicted interpretation time and a corresponding actual interpretation time for each examination; obtaining a plurality of data inputs associated with each medical image examination and/or each subject of said medical image examination, wherein data points of the plurality of data inputs represent examination variables affecting interpretation time; inputting the plurality of data inputs associated with the medical image examinations and/or the subjects, respectively, to the trained artificial intelligence algorithm; estimating corresponding optimized predicted interpretation times using the trained artificial intelligence algorithm based on said plurality of data inputs; providing said predicted interpretation times to a user of a clinical management system performing interpretations of the medical image examinations; obtaining available time of the user for image interpretation; selecting a combination of medical image examinations waiting for interpretation having a cumulative predicted interpretation time less than the user's available time; and presenting a list of the selected combination of medical image examinations to the user for the user to perform the interpretations of the medical image examinations.
2 . The method of claim 1 , wherein the artificial intelligence algorithm further provides a confidence level for each of the predicted interpretation times.
3 . The method of claim 1 , wherein the artificial intelligence algorithm ranks the predicted interpretation times for the medical image examinations and highlights a longest predicted interpretation time.
4 . The method of claim 1 , wherein the plurality of data inputs comprises: at least one of body mass index of the subject of the medical image examination, patient age, or patient gender, at least one of type of medical image study, previous image modalities available, disease type, history of diastolic dysfunction, presence of atrial fibrillation, or presence of coronary artery disease, and at least one of type of imaging modality, number of archived images or loops, or exam type.
5 . The method of claim 1 , wherein the plurality of data inputs comprises each of: body mass index of the subject of the medical image examination, type of medical image study, patient age, patient gender, previous image modalities available, disease type, history of diastolic dysfunction, presence of atrial fibrillation, presence of coronary artery disease, type of imaging modality, number of archived images or loops, sonographer's notes, and exam type.
6 . The method of claim 1 , wherein the reward function is updated using a Bellman's equation for reinforcement learning.
7 . The method of claim 1 , further comprising: obtaining actual interpretation time for each medical image examination, wherein the artificial intelligence algorithm uses the actual interpretation time and the plurality of data inputs for reinforcement learning by further minimizing the reward function to further optimize the artificial intelligence algorithm.
8 . The method of claim 1 , further comprising: providing a search function to the user that triggers the artificial intelligence algorithm to predict an interpretation time for two or more selected imaging types using a mapping function and state variables X.
9 . A clinical management system configured to optimize prediction of interpretation times for medical image examinations of subjects, the clinical management system comprising a processor operably connected to a non-transitory memory, the memory having encoded thereon machine-readable program code that, when executed by the processor, causes the processor to: train a regression-based, artificial intelligence algorithm by obtaining training data inputs and corresponding actual interpretation times for a plurality of examinations, training the artificial intelligence algorithm to map states defined by the training data inputs to corresponding predicted interpretation times using a reinforcement learning approach, and minimizing a reward function to optimize the artificial intelligence algorithm to improve the predicted interpretation times over time, wherein the reward function comprises an absolute value of a difference between the predicted interpretation time and a corresponding actual interpretation time for each examination; obtain a plurality of data inputs associated with each medical image examination and/or each subject of said medical image examination; input the plurality of data inputs associated with the medical image examinations and/or the subjects, respectively, to the trained artificial intelligence algorithm; estimate optimized predicted interpretation times using the trained artificial intelligence algorithm based on said plurality of data inputs; display said predicted interpretation times to a user of the clinical management system performing interpretations of the medical image examinations; obtain available time of the user for image interpretation; select a combination of medical image examinations waiting for interpretation having a cumulative predicted interpretation time less than the user's available time; and present a list of the selected combination of medical image examinations to the user for the user to perform the interpretations of the medical image examinations.
10 . The clinical management system of claim 9 , wherein the predicted interpretation times are presented in a table of medical imaging examinations awaiting interpretation.
11 . The clinical management system of claim 9 , wherein the machine-readable program code, when executed by the processor, further causes the processor to: provide a search function to the user that triggers the artificial intelligence algorithm to predict an interpretation time for two or more selected imaging types using a mapping function and state variables X.
12 . The clinical management system of claim 9 , wherein the artificial intelligence algorithm further automatically provides confidence levels for the predicted interpretation times, respectively, and the machine-readable program code, when executed by the processor, further causes the processor to: provide said confidence levels to the user of the clinical management system performing interpretations of the medical image examinations.
13 . A machine-readable storage media having encoded thereon program code for optimizing prediction of medical imaging interpretation time that, when executed by a processor, causes the processor to: train a regression-based, artificial intelligence algorithm by obtaining training data inputs and corresponding actual interpretation times for a plurality of examinations, training the artificial intelligence algorithm to map states defined by the training data inputs to corresponding predicted interpretation times using a reinforcement learning approach, and minimizing a reward function to optimize the artificial intelligence algorithm to improve the predicted interpretation times over time, wherein the reward function comprises an absolute value of a difference between the predicted interpretation time and a corresponding actual interpretation time for each examination; obtain a plurality of data inputs associated with each medical image examination and/or each subject of said medical image examination, wherein data points of the plurality of data inputs represent parameters affecting interpretation time; input the plurality of data inputs associated with the medical image examinations and/or the subjects, respectively, to the trained artificial intelligence algorithm; estimate corresponding optimized predicted interpretation times using the trained artificial intelligence algorithm based on said plurality of data inputs; cause said predicted interpretation times to be displayed to a user of a clinical management system performing interpretations of the medical image examinations; obtain available time of the user for image interpretation; select a combination of medical image examinations waiting for interpretation having a cumulative predicted interpretation time less than the user's available time; and present a list of the selected combination of medical image examinations to the user for the user to perform the interpretations of the medical image examinations.
14 . The machine-readable storage media of claim 13 , wherein the artificial intelligence algorithm further provides confidence levels for the predicted interpretation times, respectively.
15 . The machine-readable storage media of claim 13 , wherein the artificial intelligence algorithm ranks the predicted interpretation times for medical image examinations and highlights a longest predicted interpretation time.
16 . The machine-readable storage media of claim 13 , wherein the plurality of data inputs comprises: at least one of body mass index of the subject of the medical image examination, patient age, or patient gender, at least one of type of medical image study, previous image modalities available, disease type, history of diastolic dysfunction, presence of atrial fibrillation, or presence of coronary artery disease, and at least one of type of imaging modality, number of archived images or loops, or exam type.
17 . The machine-readable storage media of claim 13 , wherein the plurality of data inputs comprises each of: body mass index of the subject of the medical image examination, type of medical image study, patient age, patient gender, previous image modalities available, disease type, history of diastolic dysfunction, presence of atrial fibrillation, presence of coronary artery disease, type of imaging modality, number of archived images or loops, sonographer's notes, and exam type.

Description

CROSS-REFERENCE TO PRIOR APPLICATIONS This application is the U.S. National Phase application under 35 U.S.C. § 371 of International Application No. PCT/EP2022/050928, filed on Jan. 18, 2022, which claims the benefit of U.S. Provisional Patent Application No. 63/141,995, filed on Jan. 27, 2021. These applications are hereby incorporated by reference herein. FIELD OF THE INVENTION The invention relates to the field of medical image interpretation workflow and more particularly to predicting medical image interpretation times to enhance workflow, productivity and resource planning. BACKGROUND Due to an increased cardiac imaging workflow and shortage of medical experts worldwide, Echocardiogram or Echo Lab directors are under pressure to improve efficiency. Clinical management systems, such as Intellispace Cardiovascular (ISCV) from Koninklijke Philips N.V. of Eindhoven, the Netherlands, facilitate workflow management of cardiac imaging by managing images, information, and workflow and providing workflow visibility to cardiologists, lab directors, and other medical personnel at multiple locations and in multiple data configurations. An integral part of improving cardiac imaging workflow efficiency is obtaining quick turnaround times (TOT) for image interpretation, without compromising the quality of health outcomes. Currently, in most hospital settings, due to variations of clinical situations for different patients, cardiologists do not know what interpretation time to expect in advance for each image interpretation. The interpretation time can vary from 5 minutes to 50 minutes depending on the disease type, clinical features, and level of complexity of the individual image interpretation. Interpretation time uncertainty can reduce workflow efficiency. The assignment of cardiologists to exams depends on the hospital workflow. Often hospitals have attending cardiologists, assigned by day of the week, for example. Many cardiologists process their examination worklists following the first-in first-out principle. As exams arrive for image interpretation, cardiologists pick them to read. If there is any urgency, the sonographer may call the attending cardiologist to point out a specific exam that needs to be completed with higher priority. It is also common for a cardiologist to select less complex exams or exams that the cardiologist feels more confident with to read first. Neither of these workflows or resource planning methods is optimized for productivity or health outcome quality, given the uncertainty of about the time needed for interpretation of each exam. It is also common for image interpretation to be provided on a service level agreement, where final imaging reports must be completed within a specified TOT. Managing the contracted TOT is difficult when the interpretation time is unknown. There are existing workflow tools such as the worklist orchestrator management solution provided by Carestream HCIS (as part of the Clinical Collaboration Platform's architecture) that provide adaptive worklist prioritization to balance the radiologist's workload and maximize the probability that an exam be read within its service level agreement. The prioritization feature is specifically designed to show radiologists the time remaining before breaching the service level agreement. SUMMARY The inventors of the present invention have realized that accurately predicting interpretation time for medical images and making these predictions available to medical personnel can improve efficiency and enhance workflow management. Accordingly, the present invention provides a method, system, and program product for predicting interpretation time for medical images and providing these predictions to medical personnel through a clinical management system for use in workflow management. With the ratio of examination volume to the number of available cardiologists increasing, reading efficiency and the time spent by cardiologists for image interpretation to provide the final report plays a pivotal role in enabling the hospital to potentially accept more patients. Therefore, having prior knowledge about the interpretation time gives flexibility to the cardiologist to manage their time more efficiently and contribute to quicker TOT, enabling the hospital to potentially take more patients. Also, prior knowledge of interpretation time allows the hospital and the cardiologist to manage the workflow to complete the interpretation for each exam within a timeframe that supports the quality of health outcomes. According to a first aspect of the present invention, a method is provided for predicting an interpretation time for a medical image examination of a subject comprising one or more medical images. The method starts by obtaining a plurality of data inputs, where the data inputs are associated with the medical image examination and/or the subject of said medical image examination, and the data points represent parameters affecting the in